Apex

TYPE

Packaging Brand System, AI Research

AI as a research partner, not the designer.

APEX is a marine-protein snack brand built to test one question: where can AI genuinely speed up design research and prototyping, and where does the designer still have to take over? AI mapped the market and rendered ideas in minutes. The interviews, the insight and every final design decision stayed human.

01 The question

01 The question

Most conversations about AI in design ask whether it will replace designers. I wanted a more useful answer: at which steps of a real brand project does it actually help? Over 14 weeks, our team shared one research workflow and each of us built a brand from it. APEX is mine. Every step below is marked human-led, AI-assisted or both.

Research workflow

Human-ledAI-assistedHuman + AI

Stage 01

CONTEXT & SYSTEMS

01Human-led

Assumptions & Framing

Uncovering the assumptions in the industry, pinpoint key considerations, and question the status quo.

02AI-assisted

Competition & Landscape

Who is already on the shelf, and what they promise vs. what they deliver.

03AI-assisted

The Ecosystem

The invisible force that shape a choice before any product does.

04AI-assisted

STEEPx Analysis

Structural shifts, trends, and opportunities in each categories.

↓From desk research to people↓

Stage 02

RESEARCH & SYNTHESIS

05Human-led

ICP & Interview Plan

Finding out “who would buy this?”, and approaching real life consumers for inputs.

06Human-led

Interview & Insights

Real people, real routines: pains, needs, and opportunities in everyday life.

07AI-assisted

AI Personas

Each profile is simulated in an AI agent to give a larger picture.

08Human + AI

Synthesis & Strategy

The market competition, interview insights, and AI agent outputs all framed into one brand.

02 Research at speed

02 Research at speed

Research was where AI saved the most time. We planned a research agent with Claude, gathered market data with Gemini and summarized it with ChatGPT, then built the agent in Lovable so the whole team could use it. Fed with interviews, survey stats and trend research, it mapped competitors, pain points and trends, and pointed at a gap: health snacks are dominated by sweet, overly processed products, and reviews kept calling protein bars “waxy”.

How the research agent works

Human-ledAI-assistedHuman + AI
01 Input

Interviews

Survey Stats

Trend Research

Insights

Fed in by us
02 Synthesis

Synthesis & Summary

  • Consolidate points
  • Key insights
Agent working
03 Analysis

Interview Analysis

  • Pain points / tensions
  • Goals / needs
  • Habits / behaviors

Opportunities

Competitive Intelligence

  • Main + future competitors
  • Price + market share
  • Channels
  • Audience
  • Promises / claims

Customer Perception Gaps

Trend Analysis

STEEPx Analysis

Agent output
04 Output

Product Design Suggestions

  • Functions / offerings
  • Touchpoints
  • Brand language + promises
  • Packaging, color, material
Judged by us
↺LoopThe product idea goes back into the AI personas, their feedback becomes new data, and the cycle runs again.
Explore

Three Assistants, Three Jobs

  • Claude: strategic framework
  • Gemini: research + data
  • ChatGPT: summarizing + sorting
Test

A First Working Agent

  • Claude + Google AI Studio
  • Competition, interviews, synthesis
Build

Rebuilt in Lovable

  • Easier for the team to share
  • Refined + polished over many prompts

03 The human insight

03 The human insight

What the agent couldn’t do was talk to people, so I did. The athletes I interviewed had “tried them all” and still preferred “just eating real food”. Savory was what they craved. The insight came from those conversations, not the agent: athletes like Nathan Wang, 28 and training six days a week, already snack on dried squid, shrimp and fish. High-protein, minimally processed, and nobody had built a performance brand around them.

04 Real vs AI interviews

04 Real vs AI interviews

To see a bigger picture, each persona was also simulated in an AI agent and interviewed. Side by side, the difference was clear: AI personas were complete and expressive, and great for journey mapping, but templated, and they mostly echoed our own assumptions. Only the real interviews turned up something unexpected.

Real interviews vs AI personas

Human-ledAI-assistedHuman + AI
HumanHuman-led

Real Interviews

Strengths
  • More authentic
  • More specific
  • Unexpected findings
  • New opportunity areas
  • Stronger evidence
Weaknesses
  • Narrative often incomplete
  • Weaker character profile
  • Emotions less explicit
  • Not ready for persona writing
AIAI-assisted

AI Personas

Strengths
  • Complete character profile
  • Richer scenarios
  • Stronger emotional expression
  • Great for journey mapping
  • Fills in usage details
Weaknesses
  • Less authentic
  • Can feel templated
  • Reinforces assumptions
  • No truly unexpected findings
AI-assisted

When information is limited, AI helps expand thinking. When it’s complex, it helps organize it.

Human-led

The designer controls the overall framework and makes the judgment calls throughout.

05 Rapid prototyping

05 Rapid prototyping

With the direction set by research, AI became a prototyping tool. I fed Midjourney a moodboard to explore packaging, gave Luma a brand overview (product lineup, audience, moodboards) to draft brand guidelines, and generated food photography with Luma and Gemini. Ideas that would take days to mock up took minutes, so I could see the finish line early and drop weak directions fast.

06 The designer’s call

06 The designer’s call

The refining was mine. I took the brand guidelines into Illustrator and reworked them by hand, and designed the packaging panels there before feeding them back into AI only to render them. The system is three series, numbered like equipment: 01 squid chips, 02 shrimp bar, 03 fish jerky. Every final call came from my own judgment, with AI as the starting ground.

07 In the world

07 In the world

In stores and in use, APEX makes ingredients people already trust feel like gear. Tagline: Unlock your potential.

08 What I learned

08 What I learned

AI was best at two things: turning research into a starting direction, and prototyping fast. It was weakest at what matters most, understanding people and deciding what’s right. Used that way, it doesn’t replace the designer. It gives the designer more time to design.

Next

Pasadena — Los Angeles

© 2026 Naomi Tai